Fake me: face aligment via keypoint extraction for mapping expressions

Garriga Peguero, Marc

Fake me: face aligment via keypoint extraction for mapping expressions

License

Citation

Garriga Peguero, M. Fake me: face aligment via keypoint extraction for mapping expressions. 2014. handle: http://hdl.handle.net/10230/22904

This citation was generated automatically.

Abstract

The aim of this TFG is the conception and implementation of a method known in Computer Vision as Face Transfer. Given two faces, the objective of this method proposed is to be able to transfer of expressions and facial features from a face to another. This means, being able to apply a mapping that allows us to animate face or avatar of any character through the movement of another face, for instance, our face. There are many ways to map the information between two faces. In our case, we propose to divide the faces into triangular regions, through featuring points detection previously defined. Once we got the regions, we can transfer the correspondent textures using barycentric coordinates to map and a technique known as Multiband Blending to merge both face textures in a smooth way. In this TFG we have explored two scenarios in the application of Face Transfer technique. First one is a texture mapping of an image over another in movement along a video sequence. The second one is the inverse situation, an avatar animation through the facial expressions captured in a video sequence.